Bayesian estimation problem help

Estimation ML/Bayesian estimation
1

image
image1621×1206 114 KB

est4_udiag3.eps (76.9 KB)
est4_udiag4.eps (55.5 KB)
est4_udiag1.eps (78.7 KB)
est4_udiag2.eps (76.6 KB)
est4_CheckPlots2.eps (19.3 KB)
est4_CheckPlots1.eps (47.0 KB)
est4_PriorsAndPosteriors1.eps (142.8 KB)
est4_PriorsAndPosteriors2.eps (39.8 KB)

varobs C_cycle Deb_cycle Pai Q_cycle R Y_cycle;
identification(advanced=1);

estimated_params;
%rho_vh1, beta_pdf, 0.45, 0.1;
rho_bita1, beta_pdf, 0.45, 0.1;
rho_Zz, beta_pdf, 0.45, 0.1;
rho_Ah, beta_pdf, 0.45, 0.1;
rho_r, beta_pdf, 0.45, 0.1;
%rho_ltv, beta_pdf, 0.45, 0.1;
%rho_zibenyueshu, beta_pdf, 0.45, 0.1;
%rho_up, beta_pdf, 0.45, 0.1;
stderr e_bita1, inv_gamma_pdf, 0.001, 0.01;
stderr e_vh1, inv_gamma_pdf, 0.001, 0.01;
stderr EZz, inv_gamma_pdf, 0.001, 0.01;
stderr E_Ah, inv_gamma_pdf, 0.001, 0.01;
stderr Emp, inv_gamma_pdf, 0.001, 0.01;
stderr eps_ltv, inv_gamma_pdf, 0.001, 0.01;
%stderr e_zibenyueshu, inv_gamma_pdf, 0.001, 0.01;
stderr Eup, inv_gamma_pdf, 0.001, 0.01;
end;
estimation(datafile = mydata_est_cycle,
bayesian_irf,irf=120,
conf_sig=0.95,
smoother,
mh_jscale=0.33,
mode_compute=6,
mode_check,
presample=4,
prior_trunc=1e-10,
mh_replic=500000,
moments_varendo,
conditional_variance_decomposition=[1,4,8,20],
mh_nblocks=2,
lik_init=1,
plot_priors=1) Y W1_c W2_c W1_h W2_h C C1 C2 Q L1 L2 H1 H2 K Pai Deb_inc Newloan Newloan2 B2 IK_c IK_h;
shock_decomposition C_cycle Deb_cycle Pai IK_c_cycle Q_cycle R Y_cycle;

2

Without actual codes it is impossible to tell. But as a first step, you should compare means and standard deviations from a simulated version of your model with what you consider sensible parameters to the ones of the data. That often reveals the source of discrepancies.

3

ee1.mod (28.1 KB)
mydata_est_cycle.mat (72.0 KB)
Thank you very much for your reply. I have uploaded my estimation code and dataset. My model incorporates real estate producers and non-real estate producers, alongside a long-term debt amortization mechanism, which makes it relatively complex.

4

Comparing moments at the prior mean with the actual data moments shows that the data is about 10 times more volatile than the model. That may be part of the issue.

5

Thank you very much for your feedback. I initially set a relatively low prior mean to let the data dominate the estimation, and I will try raising it next.

I also have a theoretical question. I aim to simulate an economic recession driven by aggregate demand shocks. However, I observe that when the discount factor rises, households tend to save more. This increases the supply of investable funds and pushes up investment, so the economy fails to experience the severe recession as I expected.

My intended transmission mechanism is as follows: a drop in household consumption reduces firms’ revenues. Firms then cut employment and capital investment, which ultimately leads to a decline in total output. Could you offer some advice on how to properly simulate a recession caused by aggregate demand shocks?

6

What you describe sound like you need more/stronger frictions like investment adjustment costs and stickier prices.